Berlin 2024 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

DY: Fachverband Dynamik und Statistische Physik

DY 42: Focus Session: Computing with Dynamical Systems: New Perspectives on Reservoirs and Applications I – Fundamentals

DY 42.2: Vortrag

Donnerstag, 21. März 2024, 10:00–10:15, BH-N 243

Harnessing multistability: Expanding the capabilities of reservoir computers via multifunctionality — •Andrew Flynn¹, Vassilios Tsachouridis², and Andreas Amann¹ — ¹School of Mathematical Sciences, University College Cork, Cork, IrelandIreland — ²Collins Aerospace Applied Research & Technology, Cork, Ireland

Multifunctionality describes a neural network's ability to harness multistability in order to perform various tasks without altering its network properties. In this talk we demonstrate the advantages of extending multifunctionality to the domain of artificial neural networks (ANNs). Multifunctionality unlocks several new machine learning application for ANNs such as: data-driven modelling of multistability, generating chaotic itinerancy, novel memory recall techniques, and reconstructing transitions present in the epileptic brain. We outline how multifunctionality has so far been realised in an artificial setting with a reservoir computer (RC), a dynamical system in the form of an ANN. We employ generalised synchronisation to describe how to train a RC to achieve multifunctionality and also explore some of the challenges involved in realising multifunctionality. Our results not only illuminate the exotic dynamics and exciting applications of multifunctional RCs but also highlight the importance of a dynamics-driven approach when training ANNs to display a broader level of intelligence by performing multiple tasks without compromising on explainability.

Keywords: Reservoir Computer; Multistability; Neural Network; Chaotic Itinerancy; Generalised Synchronisation